The relatively high power conversion efficiency (PCE)1 of perovskite solar cells (PSCs) combined with their potential for low-cost production2 and their outstanding opto-electronic properties such as band-gap tuneability3, long charge diffusion length4, low recombination rates5, and photon recycling6,7, would make these devices ready for the PV market, although long-term stability remains a concern8. PSCs degradation can take place in the light-absorbing perovskite layer and/or in any intermediate layers, which can degrade due to their intrinsic structural instability and/or due to external factors, such as oxygen, moisture, heat, electrical bias, and mechanical stress9. Research into the degradation mechanisms of PSCs has so far predominantly focussed on regular n-i-p architectures10. Inverted p-i-n devices can potentially outclass the n-i-p stack, both in terms of efficiency and stability, provided that stable interlayer materials can be identified8. In this work, we explore the degradation kinetics of unencapsulated inverted p-i-n PSCs employing the benchmark CH3NH3PbI3 and CH3NH3PbI3−xClx, and a more thermally durable alternative FA0.83Cs0.17Pb(I0.8Br0.2)3 perovskites as the photoactive layers11,12.

Recently, we investigated the degradation kinetics of unencapsulated regular CH3NH3PbI3−xClx (MAPIC) PSCs under continuous illumination in dry N2 (stabilization phase) and N2:O2 (stress phase) atmospheres13. Current-voltage (IV) sweeps, transient photocurrent (TPC) and transient photovoltage (TPV) measurements were continuously and sequentially acquired in-situ and in-operando. During the stress phase the PCE was exponentially lost over time due to the emergence of a space-charge within the device that impeded charge extraction and accelerated photo-oxidation of the perovskite layer13. Here, we use the same setup to age MAPIC, CH3NH3PbI3 (MAPI) and FA0.83Cs0.17Pb(I0.8Br0.2)3 (mixed-cation) PSCs. The intrinsic stability of MAPI is poor due to the volatility of the methylammonium (MA) cation14. As MA sublimates, the perovskite converts into PbI2-rich domains that lower the efficiency of charge generation and impede charge transport between perovskite grains, thus affecting the open-circuit voltage (Voc) and the short-circuit current (Jsc)9. To overcome these issues, more structurally stable perovskites have been obtained by replacing the MA cation with complex cation mixtures15,16,17,18. The caesium/formamidinium (Cs/FA) combination has been used to fabricate structurally stable and band-gap tuneable FA0.83Cs0.17Pb(IxBr1−x)3 regular PSCs with relatively high PCEs12,19,20. Here we use the mixed-cation devices to provide a point of comparison between PSCs with active layers of differing intrinsic stability.

Results and Discussion

Evolution of Current-Voltage Figures-of-Merit During Ageing

In Fig. 1 we show the evolution of the normalized figures-of-merit (FOM) extracted from reverse IV sweeps (Figures S1S3) of the three inverted devices stressed under continuous simulated solar illumination (AM 1.5 G) in dry N2 and in dry N2 (99%): O2 (1%) atmospheres. All devices discussed here have the architecture FTO/PEDOT:PSS/Poly-TPD/perovskite/PCBM/BCP/Au (see SI for Materials and Methods). Note that here data is normalized twice to facilitate a comparison between the relative changes in the metrics of the three devices during both phases (see also Figures S4S7). Such a device structure results in a negligible hysteresis (Figures S8S10) compared to analogous regular n-i-p devices13, due to the good charge extraction properties of PCBM, and presumably fewer defects responsible for charge recombination at the perovskite charge extraction layer interface21.

Figure 1
figure 1

Evolution of the figures-of-merit (FOM) of inverted MAPIC, MAPI, and mixed-cation PSCs under continuous illumination and dry N2 (Time < 0) and dry N2 (99%) with O2 (1%) (Time > 0). FOM are extracted from the IV reverse sweeps (from 1 V to 0 V). Metrics during stabilization (Time < 0) and stress (Time > 0) phases are normalized to the first recorded value of the reverse metrics during stabilization and stress phases, respectively. Normalized open-circuit voltage Voc (a), short-circuit current Jsc (b), fill factor FF (c) and power conversion efficiency PCE (d).

During the stabilization phase (Time < 0) all devices undergo a reduction in PCE, with the MAPI PSC experiencing total failure within 20 hours. The loss in the PCE of the MAPIC and mixed-cation PSCs is mainly due to a reduction in Jsc, however for the MAPI device, the Voc also reduces. The superior stability of MAPIC over MAPI during the stabilization phase could be an effect of PbCl2 in the precursor solution resulting in a perovskite layer with improved morphology and/or lower defect density22,23, although the exact mechanism(s) for stability enhancement are not fully understood. During the stress phase (Time > 0) the MAPIC PSC completely degrades to ~5% of its initial PCE over 20 hours whilst the mixed-cation device retains ~70% of its initial PCE at Time = 0. The Voc of the mixed-cation PSC remains constant throughout both phases, suggesting that the perovskite remains stable and is not apparently affected by halide segregation. The MAPIC device turned into yellow colour at the end of the stress phase, consistent with the known mechanism for generation and reaction of superoxide (O2), which subsequently decomposes the methylammonium halide within the perovskite crystal24. In general, for all the devices, most of the losses in the PCE are due to losses in the Jsc. Therefore, we postulate that photo-oxidization, or degradation of the charge extraction layers25, or their interface with the perovskite could be playing a role with reducing the charge extraction efficiency.

Evolution of Transient Photocurrent During Ageing

To explore in detail the photocurrent loss mechanisms we consider the evolution in TPC traces measured in sequence with the IV scans during the stabilization and stress phases (see Figures S11S13). From these traces we identify five types of photocurrent behaviour, which are represented in Fig. 2a.

Figure 2
figure 2

Stability kinetics extracted from transient photocurrent (TPC) measurements of MAPIC, MAPI, and mixed-cation PSCs under continuous illumination and dry N2 (Time < 0) and dry N2 (99%) with O2 (1%) (Time > 0). (a) Offset plot of types of behaviour for normalized TPC curves. (b) Extracted charge obtained by integrating the TPC curves after the LED is switched off (Time > 0.4 ms). Different colours indicate a curve type (transition regions are indicated with a colour blend). Different line thickness indicate data from different solar cells: MAPI (dotted line), MAPIC (thin line), Mixed-cation (thick line).

Slow components (dominant in type 3 and 4) are typically attributed to charge trapping/de-trapping and recombination processes, while fast transients (dominant in type 1, 2 and 5) are compatible with timescales associated with charge carrier transport26,27,28. The TPC dataset was clustered with a pattern recognition neural network (PRNN), which is a software-based computing system that works similarly to biological nervous systems29,30,31,32. Once trained to recognize certain patterns, PRNNs can output fuzzy or intermediate answers. Here a PRNN (Figure S14) is trained with the TPC dataset shown in Fig. 2a to provide a qualitative description of the TPC shape evolution during ageing. In Fig. 2b we plot the extracted charge from the photocurrent decay transients and indicate the TPC curve types evolution during ageing. At the beginning of the stabilization phase all devices behave according to type 1 with a fast transient when the LED is switched on/off, which is indicative of the relatively clean and efficient photocurrent generation behaviour of the as-fabricated PSCs13. Continuous operation in N2 induces changes in the TPC shape for all devices. The TPC of the MAPIC device transitions from type 1 to type 2 after ~6 hours of ageing. The photocurrent overshoot in type 2, observed in the first few μs of PSC illumination, may be attributed to the rapid formation of a transient diffusion gradient that enhances charge carrier recombination (reduces the photocurrent) before fading27,33. The TPC of the MAPI PSC transitions from type 1 to type 4 after only ~3 hours of ageing, during which time the extracted charge from the photocurrent decays progressively reduces until the solar cell stops working. The TPC characteristics of the mixed-cation PSC immediately transitions from type 1 to type 5 after ~1 hour of ageing and maintains this behaviour until the last ~5 hours of the stress phase, when it goes back to type 1. Throughout ageing of the mixed-cation PSCs, the extracted charge experiences a negligible drop. In the MAPIC PSC, after ~1.5 hour exposure to oxygen the TPC transitions from type 2 to type 5 with a continuous decrease in the extracted charge. As the TPC traces further evolve from type 5 to type 3 the decay signal becomes negative, which is indicative of charge injection into the cell34. This observation and the photocurrent decay during the LED ‘on’ period could be explained by enhanced trap-assisted recombination and reduced charge de-trapping rate mechanisms35. While the charge density within the device increases due to continued photoexcitation, the competition between charge recombination and charge extraction in the PSC favours the former process to an extent that the steady-state photocurrent decreases. An increase in charge density within the PSC may also result in a space-charge that opposes the built-in field, resulting in a lower charge extraction efficiency13. For the mixed-cation PSC, the fact that the TPC shape does not seem to be influenced by the presence of oxygen indicates the superior stability of this device. We also observe that TPC type 4, seen during periods of severe photo-degradation, and type 3, which is dominant during the stress period, are both characterized by slow photocurrent decay transients (prolonged charge de-trapping and injection), compared to the other curves.

Evolution of Transient Photovoltage During Ageing

TPV measurements (see Figures S17S22) provide complementary information on the generation/recombination kinetics of photo-generated charges in the small perturbation regime34. In our degraded solar cells the TPV decays are best fitted with a double exponential function (see Figures S23S28)22,36. In Fig. 3a we show that for the MAPIC device the fast time constant (T2) dominates during stabilization (a2 > a1). During the stress phase however the slow time constant (T1) increases and becomes dominant (a1 > a2) within ~5 hours before stabilizing. Although the origin of the slow and fast components are still under debate13,36, we note that the double exponential behaviour is indicative of two populations of carriers that independently recombine22. In Fig. 3b we show the evolution of the slow time constant T1 versus Voc during the stress phase. This trend is compared to the ideal behaviour of the same device prior to ageing obtained by measuring TPVs under different light intensities. The T1 vs Voc trend during the stress phase is non-linear with remarkably higher time constants compared to the ideal behaviour, suggesting that the time constants measured during the stress phase are likely to originate from trapped charges within the perovskite layer rather than free carriers. However, for the mixed-cation the dominant time constant is significantly lower (~1–2 μs) and remains stable throughout ageing (Figures S26S28). This indicates that traps are not being generated in the perovskite layer and that the observed degradation might be due to degrading interlayers reducing current extraction and increasing the series resistance (Figures S29S30).

Figure 3
figure 3

Stability kinetics extracted from transient photovoltage (TPV) measurements of MAPIC, MAPI, and mixed-cation PSCs under continuous illumination and dry N2 (Time < 0) and dry N2 (99%) with O2 (1%) (Time > 0). (a) Extracted time and amplitude constants from the double exponential fits of the TPV transient decays (a1 × exp(−x/T1) + a2 × exp(−x/T2), where x is the time (μs)) for the MAPIC device. (b) Slow photovoltage decay time constant (T1) versus Voc for the MAPIC device during the stress phase compared to reference values (black curve).

Light Dependant Current-Voltage

To further understand the recombination dynamics of the solar cells under stress conditions, we measured IV sweeps under variable light intensities (1–100 mW cm−2 AM 1.5 G) before stabilization, at the end of stabilization, and at the end of the stress phase. The Voc versus the natural logarithm of the light intensity shows a linear behaviour (Figure S31) and from its slope nkT/q we can extract the ideality factor n (Fig. 4)37,38. For the mixed-cation PSC n ≈ 1 throughout ageing indicates bimolecular charge recombination39. For the MAPIC PSC the progressive increase of n from ~1.66 to 2.53 during the stabilization phase indicates an increase in Shockley-Reed-Hall trap-based recombination. Further, we examine the power law dependence of Jsc with light intensity (Jsc Iα) (Figure S32). The fitted alpha parameter (Fig. 4) reduces throughout ageing for both MAPIC and the mixed-cation PSCs indicating the possible presence of trapped charges within the perovskite layer40.

Figure 4
figure 4

Ideality (n) and alpha factor extracted from Voc and Jsc vs Light Intensity, respectively. Measurements are performed before the beginning of the stabilization phase, at the end of the stabilization phase, and at the end of the stress phase. Full lines represent the extracted ideality factor, as indicated by the black arrow. Dotted lines represent the extracted alpha factor, as indicated by the black dotted arrow. Vertical lines indicate the 95% confidence intervals for n and alpha extracted from the fits.


In summary, we investigated the operational stability kinetics of unencapsulated CH3NH3PbI3, CH3NH3PbI3−xClx and FA0.83Cs0.17Pb(I0.8Br0.2)3 inverted perovskite solar cells in the presence of light and dry oxygen using in-situ and in-operando IV, TPC and TPV measurements. We confirm the superior stability of the mixed-cation PSCs compared to the benchmark PSCs. The observed light- and oxygen-induced degradation in the MAI-based solar cells occurs over shorter timescales than the mixed-cation devices, and is dominated by a loss in photocurrent and charge extraction efficiency. We interpret this to the generation of electron traps, resulting in long-lived trapped charge and the build-up of space-charge within the perovskite absorber layer. Our findings provide important insights towards understanding the operation of perovskite solar cells, and suggest that focussing on mitigating trap generation during ageing will lead to further improvements in perovskite solar cell operation.